CN117633588A - Pipeline leakage positioning method based on spectrum weighting and residual convolution neural network - Google Patents
Pipeline leakage positioning method based on spectrum weighting and residual convolution neural network Download PDFInfo
- Publication number
- CN117633588A CN117633588A CN202311579219.7A CN202311579219A CN117633588A CN 117633588 A CN117633588 A CN 117633588A CN 202311579219 A CN202311579219 A CN 202311579219A CN 117633588 A CN117633588 A CN 117633588A
- Authority
- CN
- China
- Prior art keywords
- leakage
- neural network
- pipeline
- frequency
- layer
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000001228 spectrum Methods 0.000 title claims abstract description 56
- 238000000034 method Methods 0.000 title claims abstract description 45
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 17
- 230000003287 optical effect Effects 0.000 claims abstract description 28
- 238000012545 processing Methods 0.000 claims abstract description 22
- 238000003062 neural network model Methods 0.000 claims abstract description 17
- 238000012549 training Methods 0.000 claims description 39
- 238000013527 convolutional neural network Methods 0.000 claims description 36
- 238000012360 testing method Methods 0.000 claims description 20
- 230000008569 process Effects 0.000 claims description 14
- 238000005070 sampling Methods 0.000 claims description 9
- 230000004913 activation Effects 0.000 claims description 5
- 239000000284 extract Substances 0.000 claims description 5
- 238000004590 computer program Methods 0.000 claims description 4
- 230000003595 spectral effect Effects 0.000 claims description 4
- 230000004807 localization Effects 0.000 claims description 2
- 230000006870 function Effects 0.000 description 23
- VNWKTOKETHGBQD-UHFFFAOYSA-N methane Chemical compound C VNWKTOKETHGBQD-UHFFFAOYSA-N 0.000 description 8
- 238000012544 monitoring process Methods 0.000 description 7
- 230000007613 environmental effect Effects 0.000 description 4
- 239000003345 natural gas Substances 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 238000012512 characterization method Methods 0.000 description 2
- 230000000295 complement effect Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 239000007789 gas Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- 230000032683 aging Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 229910052799 carbon Inorganic materials 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000005265 energy consumption Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 238000010438 heat treatment Methods 0.000 description 1
- 210000005036 nerve Anatomy 0.000 description 1
- 210000002569 neuron Anatomy 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000010355 oscillation Effects 0.000 description 1
- 238000004088 simulation Methods 0.000 description 1
- 230000005236 sound signal Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Landscapes
- Examining Or Testing Airtightness (AREA)
Abstract
The invention discloses a pipeline leakage positioning method based on spectrum weighting and residual convolution neural network, which comprises the following steps: s1, collecting vibration signals in a grating array vibration sensing optical cable, wherein the grating array vibration sensing optical cable is paved above a pipeline in a linear paving mode; s2, extracting a main frequency in the vibration signal, and carrying out frequency spectrum weighting processing in a certain frequency range by taking the main frequency as a center to obtain a reinforced vibration signal; s3, inputting the reinforced vibration signals into a trained one-dimensional large-kernel residual convolution neural network model to obtain the pipeline leakage type probability and position the leakage position. The invention has high recognition rate for gas pipeline leakage positioning.
Description
Technical Field
The invention relates to the field of machine learning, in particular to a pipeline leakage positioning method based on a spectrum weighting and large-kernel residual convolution neural network.
Background
Natural gas is a high-quality, green and low-carbon energy source, has increasingly increased importance in global energy consumption structures, and provides opportunities for the development of the natural gas industry. The pipeline becomes the first choice mode of natural gas transportation by virtue of the advantages of large transportation capacity and low cost. However, the pipe is susceptible to leakage due to environmental, pipe aging, external forces, corrosion, and the like. The leakage of the pipeline not only causes the waste of resources, but also pollutes the environment and even causes the loss of life and property. Therefore, natural gas pipeline leakage monitoring and positioning are of great significance.
In the current common pipeline leakage monitoring and positioning method, the traditional machine learning method is mainly adopted for feature extraction and leakage positioning. However, the traditional machine learning method has the defects of complicated characteristic extraction process and lower leakage positioning accuracy.
The patent discloses a heating pipeline leakage detection method based on an AlexNet convolutional neural network, which uses wavelet analysis to carry out filtering treatment on collected data, reduces interference of pipeline noise, extracts various characteristic parameters of leakage signal characterization, constructs the AlexNet convolutional neural network according to the extracted parameters, and further judges whether leakage occurs or not, but the method lacks generalization and cannot effectively cope with more complex leakage environments.
The patent also discloses a gas pipeline leakage identification method based on a convolutional neural network, which is used for carrying out short-time Fourier transform on short-time audio signals to obtain a video image of a characterization signal, and classifying leakage images by using two-dimensional convolutional nerves to realize leakage identification.
Disclosure of Invention
Aiming at the fact that vibration signals generated during pipeline leakage are complex, collected signals are easy to be interfered by environmental noise, and leakage points are difficult to accurately position, the invention provides the pipeline leakage monitoring and positioning method and system based on the spectrum weighting and large-kernel residual convolution neural network, which can adapt to more complex leakage environments and can optimize the leakage identification process.
The technical scheme adopted by the invention is as follows:
the pipeline leakage positioning method based on the spectrum weighting and residual convolution neural network comprises the following steps:
s1, collecting vibration signals in a grating array vibration sensing optical cable, wherein the grating array vibration sensing optical cable is paved above a pipeline in a linear paving mode;
s2, extracting a main frequency in the vibration signal, and carrying out frequency spectrum weighting processing in a certain frequency range by taking the main frequency as a center to obtain a reinforced vibration signal;
s3, inputting the reinforced vibration signals into a trained one-dimensional large-kernel residual convolution neural network model to obtain the pipeline leakage type probability and position the leakage position;
the training process of the one-dimensional large-kernel residual convolution neural network model is as follows: dividing a grating array vibration sensing optical cable paved above a pipeline into a plurality of areas, and arranging leakage holes with a plurality of apertures on the pipeline, wherein leakage in different working conditions is simulated by using different pressures; collecting vibration signals generated by leakage of pipelines of the grating array vibration sensing optical cable under different pressures and different apertures; extracting the main frequency of each sensing area when the pipeline leaks, calibrating the main frequency with the largest occurrence frequency as a leakage frequency, and carrying out frequency spectrum weighting processing in a certain frequency range by taking the leakage frequency as a center to obtain a processed leakage vibration signal as a data sample set; dividing a data sample set into a training set and a testing set, training a pre-constructed one-dimensional large-core residual error convolutional neural network model by adopting the training set, and testing the performance of the one-dimensional large-core residual error convolutional neural network model by adopting the testing set.
By adopting the technical scheme, three different pressures are specifically used, two different leakage holes are used for simulating different working conditions, and a plurality of leakage columns on the pipeline are positioned.
According to the technical scheme, when the vibration signals in the data sample set are collected, firstly, the high-frequency instrument is used for collecting the vibration signals generated by leakage, and if the collected signals find that the vibration signals generated by leakage of the pipeline are low-frequency signals, the pipeline leakage signals are subjected to downsampling processing, and the sampling signals are reduced from high frequency to low frequency.
In the above technical solution, during the spectrum weighting process, the spectrum weighting process is performed in the 1Hz range with the leakage frequency as the center.
The technical scheme is connected, wherein the architecture of the one-dimensional large-kernel residual convolution neural network model is as follows:
the first layer is a stage block layer, and the leakage data firstly extracts signal characteristics through a plurality of convolution layers of the stage block layer;
the second layer is a Large Kernel layer and a Small Kernel layer, the output signal features of the first layer respectively enter the Large Kernel layer to extract the global features of the vibration signals, enter the Small Kernel layer to extract the texture features of the vibration signals, and are connected with the signal features of the first layer in a residual way after the global features and the texture features are added;
the third layer, the fourth layer and the second layer have the same structure and are used for extracting deep complex features to form multi-level feature expression;
the fifth layer is a full-connection layer, and the output signal characteristics of the fourth layer are flattened by using a linear layer;
the sixth layer is a softmax layer, and the flattened signal features are classified and output for pipeline leakage using a softmax function.
By adopting the technical scheme, the one-dimensional large-kernel residual convolution neural network model adopts a ReLU activation function, and the loss function adopts a cross entropy function.
With the above technical solution, the convolution Kernel in the Large Kernel layer is larger than the convolution Kernel in the Small Kernel layer.
By adopting the technical scheme, the loss function of the one-dimensional large-kernel residual convolution neural network model adopts a cross entropy function:
wherein m represents the number of samples participating in network training at a time, y (i) Representing the true probability that the i-th sample is of the leakage class,the i-th sample is represented as the leakage class probability calculated by model training.
The invention also provides a pipeline leakage positioning system based on the spectrum weighting and residual convolution neural network, which comprises the following steps:
the signal acquisition module is used for acquiring vibration signals in the grating array vibration sensing optical cable, and the grating array vibration sensing optical cable is paved above the pipeline in a linear paving mode;
the signal processing module is used for extracting the main frequency in the vibration signal, and carrying out frequency spectrum weighting processing in a certain frequency range by taking the main frequency as a center to obtain a reinforced vibration signal;
the identification module is used for inputting the reinforced vibration signals into a trained one-dimensional large-kernel residual convolution neural network model to obtain the pipeline leakage type probability and position the leakage position;
the training process of the one-dimensional large-kernel residual convolution neural network model is as follows: dividing a grating array vibration sensing optical cable paved above a pipeline into a plurality of areas, and arranging leakage holes with a plurality of apertures on the pipeline, wherein leakage in different working conditions is simulated by using different pressures; collecting vibration signals generated by leakage of pipelines of the grating array vibration sensing optical cable under different pressures and different apertures; extracting the main frequency of each sensing area when the pipeline leaks, calibrating the main frequency with the largest occurrence frequency as a leakage frequency, and carrying out frequency spectrum weighting processing in a certain frequency range by taking the leakage frequency as a center to obtain a processed leakage vibration signal as a data sample set; dividing a data sample set into a training set and a testing set, training a pre-constructed one-dimensional large-core residual error convolutional neural network model by adopting the training set, and testing the performance of the one-dimensional large-core residual error convolutional neural network model by adopting the testing set.
The invention also provides a computer storage medium, in which a computer program executable by a processor is stored, and the computer program executes the pipeline leakage positioning method based on the spectrum weighting and residual convolution neural network.
The invention has the beneficial effects that: the invention provides a leakage monitoring and positioning method based on spectrum weighting and large-kernel residual convolution neural network, which is used for carrying out spectrum weighting pretreatment on leakage data and enhancing leakage signal characteristics under the condition that leakage signals are not distorted; by combining the grating array vibration sensing technology with the one-dimensional large-kernel residual error convolutional neural network, the advantages of high sensitivity, high reusability and high spatial resolution of the grating array and the accurate recognition capability of the one-dimensional large-kernel residual error convolutional neural network on leakage signals complement each other, the situation that environmental noise is misjudged as leakage signals is avoided, leakage signals can be captured, leakage judgment can not occur, and the pipeline leakage positioning accuracy is greatly improved.
Further, in the sample data acquisition process of model training, vibration signals generated by pipeline leakage are acquired by using the grating array vibration sensing optical cable and the high-frequency instrument so as to avoid missing high-frequency leakage vibration signals; if the vibration signal generated by the pipeline leakage is determined to be a low-frequency signal, the sampling frequency is reduced to be low-frequency, so that redundant signals of the pipeline leakage are removed, the model training time is shortened, and the network model training speed is increased
Furthermore, the used one-dimensional large-kernel residual error convolution neural network can capture global characteristics of leakage signals through large-kernel convolution and small-kernel convolution, can also consider local information of the leakage signals, introduces a residual error structure, improves generalization capability of a model, and accordingly achieves accurate positioning of leakage monitoring.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions of the prior art, the following description will briefly explain the drawings used in the embodiments or the description of the prior art, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings can be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a flow chart of a method for locating pipeline leakage based on a spectral weighting and residual convolution neural network in accordance with an embodiment of the present invention;
FIG. 2 is a graph of a leakage signal spectrum according to an embodiment of the present invention;
FIG. 3 is a spectrum of a leakage signal after spectral weighting in accordance with an embodiment of the present invention;
FIG. 4 is a diagram of a one-dimensional large-kernel residual convolutional neural network model in accordance with an embodiment of the present invention;
FIG. 5 is a schematic diagram of overall pipe leak location in accordance with an embodiment of the present invention.
Detailed Description
The present invention will be described in further detail with reference to the drawings and examples, in order to make the objects, technical solutions and advantages of the present invention more apparent. It should be understood that the specific embodiments described herein are for purposes of illustration only and are not intended to limit the scope of the invention.
Example 1
As shown in fig. 1, the pipeline leakage positioning method based on the spectrum weighted and residual convolution neural network in the embodiment of the invention comprises the following steps:
s1, collecting vibration signals in a grating array vibration sensing optical cable, wherein the grating array vibration sensing optical cable is paved above a pipeline in a linear paving mode;
s2, extracting a main frequency in the vibration signal, and carrying out frequency spectrum weighting processing in a certain frequency range by taking the main frequency as a center to obtain a reinforced vibration signal;
s3, inputting the reinforced vibration signals into a trained one-dimensional large-kernel residual convolution neural network model to obtain the pipeline leakage type probability and position the leakage position;
the training process of the one-dimensional large-kernel residual convolution neural network model is as follows: dividing a grating array vibration sensing optical cable paved above a pipeline into a plurality of areas, and arranging leakage holes with a plurality of apertures on the pipeline, wherein leakage in different working conditions is simulated by using different pressures; collecting vibration signals generated by leakage of pipelines of the grating array vibration sensing optical cable under different pressures and different apertures; extracting the main frequency of each sensing area when the pipeline leaks, calibrating the main frequency with the largest occurrence frequency as a leakage frequency, and carrying out frequency spectrum weighting processing in a certain frequency range by taking the leakage frequency as a center to obtain a processed leakage vibration signal as a data sample set; dividing a data sample set into a training set and a testing set, training a pre-constructed one-dimensional large-core residual error convolutional neural network model by adopting the training set, and testing the performance of the one-dimensional large-core residual error convolutional neural network model by adopting the testing set.
Further, three different pressures are specifically used, two different leakage holes are used for simulating different working conditions, and a plurality of leakage columns on the pipeline are positioned. If three different pressures of 0.05Mpa,0.1Mpa and 0.2Mpa and leakage holes of 5mm and 10mm can be used for simulating leakage in different working conditions, 4 leakage columns can be arranged on a pipeline, and the grating array is used for respectively collecting leakage vibration data of each leakage column in different working conditions.
Preferably, when the vibration signals in the data sample set are collected, firstly, a high-frequency instrument is used for collecting vibration signals generated by leakage, and if the collected signals find that the vibration signals generated by the leakage of the pipeline are all low-frequency signals, the pipeline leakage signals are subjected to downsampling processing, and the sampling signals are reduced from high frequency to low frequency. If each working condition simulates leakage, multiple groups of signals are generated, which correspond to multiple grating areas respectively. The vibration signal generated by the pipeline leakage is collected by the high-frequency instrument so as to avoid missing the high-frequency signal. If the observation shows that the pipeline leakage vibration signals are mainly distributed at low frequency, the down-sampling method is used for down-sampling the high-frequency vibration signals to low frequency, so that redundant signals can be removed, and the model training time is shortened.
Preferably, the architecture of the one-dimensional large-kernel residual convolutional neural network model can be a six-layer architecture, and the architecture is specifically as follows:
the first layer is a stage block layer, and the leakage data firstly extracts signal characteristics through a plurality of convolution layers of the stage block layer;
the second layer is a Large Kernel layer and a Small Kernel layer, the output signal features of the first layer respectively enter the Large Kernel layer to extract the global features of the vibration signals, enter the Small Kernel layer to extract the texture features of the vibration signals, and are connected with the signal features of the first layer in a residual way after the global features and the texture features are added;
the third layer, the fourth layer and the second layer have the same structure and are used for extracting deep complex features to form multi-level feature expression;
the fifth layer is a full-connection layer, and the output signal characteristics of the fourth layer are flattened by using a linear layer;
the sixth layer is a softmax layer, and the flattened signal features are classified and output for pipeline leakage using a softmax function.
The one-dimensional large-kernel residual convolution neural network model adopts a ReLU activation function, and the loss function adopts a cross entropy function.
The convolution Kernel in the Large Kernel layer is larger than that in the Small Kernel layer, global features of the leakage signal can be better extracted through the Large convolution Kernel, and local features of the leakage signal can be better extracted through the Small convolution Kernel.
Preferably, the loss function of the one-dimensional large kernel residual convolutional neural network model adopts a cross entropy function:
wherein m represents the number of samples participating in network training at a time, y (i) Representing the true probability that the i-th sample is of the leakage class,the i-th sample is represented as the leakage class probability calculated by model training.
According to the embodiment, the grating array vibration sensing technology is combined with the one-dimensional large-kernel residual error convolutional neural network, so that the advantages of high sensitivity, high reusability and high spatial resolution of the grating array and the accurate recognition capability of the one-dimensional large-kernel residual error convolutional neural network on leakage signals complement each other, the situation that environmental noise is misjudged as leakage signals is avoided, leakage signals can be captured, leakage judgment cannot occur, and the pipeline leakage positioning accuracy is greatly improved.
Example 2
This embodiment is based on embodiment 1, except that a preferred one-dimensional large-kernel residual convolutional neural network model is built.
As shown in fig. 4, the specific architecture of the one-dimensional large-kernel residual convolutional neural network model constructed in this embodiment is as follows:
the first layer is a stage block layer, three convolution layers are arranged in the stage block layer, the convolution kernel size is 3*3, the number is 64,128 and 256 respectively, the step pitch is 2, and batch standardization is carried out after each convolution layer.
The second layer is a Large Kernel layer and a Small Kernel layer, three convolution layers are arranged in the Large Kernel layer, the convolution Kernel size is 41 x 41, the number is 256,512 and 1024 respectively, the step pitch is 1, and batch standardization is carried out after each convolution layer. Three convolution layers were included in the Small Kernel layer, with a convolution Kernel size of 3*3, a number of 256,512 and 1024, respectively, and a stride of 1, each of which was followed by a batch normalization. The leakage data firstly passes through a stage block layer, the characteristic signals respectively enter a Large Kernel layer and a Small Kernel layer, the characteristics of the two layers are added, and then residual connection is carried out on the characteristic signals of the upper layers.
This problem can be optimized if the direct use of large-kernel convolution reduces accuracy and the small-kernel convolution using stage block re-parameterizes. Compared with the small convolution, the large convolution has better receptive field, can capture global information, can realize the judgment of the whole signal during leakage, focuses on shape characteristics (focuses on the whole graph and the structure of the whole frequency spectrum), captures texture characteristics of a leakage signal (focuses on the details of the frequency spectrum) by utilizing the small-kernel convolution, has complex frequency spectrum of the leaked vibration signal, has a plurality of frequency characteristics, can be aliased with the leakage signal, and can more comprehensively extract the leakage data characteristics by using two convolution cores with different sizes.
A second layer: large kernel block layer: the purpose is to extract the global feature of the vibration signal generated during leakage, and capture the more macroscopic feature. small kernel block layer: the aim is to extract local features of the vibration signal. And adding the characteristic signals extracted twice, and fusing the local information and the global information.
The third layer and the fourth layer have the same structure as the second layer. And repeating the operation of the third layer and the fourth layer to form multi-level feature expression, and extracting deep complex features.
The fifth layer is a fully connected layer with 1024 neurons that flatten the features using a linear layer.
The sixth layer is a softmax layer, and the sort output is performed using the softmax function.
The activation of each layer in the network model is a ReLU activation function.
The loss function of the network model adopts a cross entropy function:
wherein m represents the number of samples participating in network training at a time, y (i) Representing the true probability that the i-th sample is of the leakage class,the ith sample is represented as leakage category probability obtained by calculation through training a model;
the one-dimensional large-kernel residual convolution neural network in the embodiment can capture global characteristics of leakage signals through large-kernel convolution and small-kernel convolution, can also consider local information of the leakage signals, introduces a residual structure, improves generalization capability of a model, and accordingly achieves accurate positioning of leakage monitoring.
Example 3
This example is based on example 1, except that this example gives a specific source of sample data and training procedure for the one-dimensional large-kernel residual convolutional neural network.
The specific acquisition and processing process of the sample data of the one-dimensional large-kernel residual convolutional neural network is as follows:
step 1: and paving the grating array vibration sensing optical cable above the pipeline in a linear paving mode. As shown in fig. 5, the grating array vibration sensing optical cable is connected to a grating array demodulator to demodulate the vibration signal acquired in real time. The grating array demodulator is communicated with the upper computer through an Ethernet port.
Step 2: collecting vibration signals generated by leakage of pipelines under different pressures and different apertures through a grating array vibration sensing optical cable;
step 3: firstly, a 1000Hz instrument is used for collecting vibration signals generated by leakage so as to avoid missing high-frequency leakage vibration signals, and the collected signals find that the vibration signals generated by the leakage of the pipeline are low-frequency signals, so that the pipeline leakage signals are subjected to downsampling treatment, and the 1000Hz downsampled pipeline leakage signals are 100Hz;
step 4: and extracting the main frequency of each sensing area when the pipeline leaks, calibrating the main frequency with the largest occurrence frequency as the leakage frequency, and carrying out frequency spectrum weighting processing in a certain frequency range by taking the leakage frequency as the center.
Step 5: dividing the leakage vibration signals obtained in the step 3 into a training set and a testing set, firstly adopting the training set to train a model, and then utilizing the testing set to test the performance of the model.
In a further preferred scheme, in the step 2, three different pressures of 0.05Mpa pressure, 0.1Mpa pressure and 0.2Mpa pressure are used, two different leakage holes of 5mm leakage holes and 10mm leakage holes are used for simulating different working conditions, and four leakage columns on a pipeline are positioned.
According to the embodiment, vibration signals generated by pipeline leakage are collected through the grating array vibration sensing optical cable and the 1000Hz instrument, the vibration signals generated by the pipeline leakage are determined to be low-frequency signals, the sampling frequency is reduced to 100Hz, redundant signals of the pipeline leakage are removed, the training speed of a network model is accelerated, the main frequency of the leakage signals is calculated, the main frequency range is subjected to frequency spectrum weighting processing, and the characteristics of the leakage signals are enhanced under the condition that the leakage signals are not distorted.
Example 4
This embodiment is based on embodiment 1, embodiment 2 or embodiment 3, and differs mainly in the spectrum weighting process of the sampled data.
In this embodiment, multiple sets of signals (e.g., 10 sets of signals) are generated during each leakage simulation under each working condition, and the collected leakage signal spectrograms respectively correspond to multiple grating areas, as shown in fig. 2. A high frequency (e.g., 1000 Hz) meter is used to collect vibration signals generated by pipe leaks so as to avoid missing high frequency signals. Through observation, the pipeline leakage vibration signals are mainly distributed at low frequency, and a down-sampling method is used for down-sampling the high-frequency vibration signals to low frequency (100 Hz) so as to remove redundant signals and shorten the model training time.
After the main frequency of each of the plurality of grating regions is calculated, the main frequency with the largest occurrence number is extracted, and the main frequency is calibrated as the secondary leakage frequency. And then, taking the main frequency as the center, and carrying out spectrum weighting processing on the plurality of grating monitoring areas by using a Hamming window function in a certain frequency range (such as left and right 1 Hz).
The hamming window weighting function is a commonly used window function that is shaped like a rectangular window with smooth transitions. It reduces boundary effects in the spectrum by attenuating oscillations across the window. The hamming window weighting function can be expressed as:
a 0 the value of (2) determines the weight of the cosine term and thus affects the shape of the whole window. When a is 0 = 0.53836, called Hamming window; when a is 0 =0.5, then it is called Hann window. The Hann window is also called raised cosine window. The Hann window can be regarded as the sum of the spectrums of 3 rectangular time windows, or the sum of 3 sin (T) type functions, and two terms in brackets are respectively shifted to the left and right by pi/T relative to the first spectral window, so that side lobes cancel each other, and high-frequency interference and energy leakage are eliminated. If will a 0 Setting to a value close to 0.53836, or more precisely 25/46, gives a Hamming window, and setting this value has the effect of creating a zero-crossing at a frequency of 5 pi/(N-1) so that the first side lobe (side lobe) of the original Hann window can be largely eliminated, yielding only a Hann window 1/5 heightSide lobes of the degree. Wherein w (N) represents the value of the Hamming window at index N, and N is the window length.
In the frequency domain weighting process, each frequency component in the spectrum is weighted by multiplying it by the hamming window value at the corresponding position. Assuming that the frequency spectrum obtained by fourier transform is X (k) and the weighted frequency spectrum is Xw (k), the calculation formula of the frequency domain weighting is as follows:
Xw(k)＝X(k)*W(k)
where X (k) represents the value of the spectrum at the frequency index k and W (k) represents the value of the hamming window at the frequency index k.
The spectrum diagram of the signal spectrum weighted in this embodiment is shown in fig. 3, and it can be seen that this embodiment further enhances the leakage signal characteristics by performing spectrum weighting processing on the main frequency range without distortion of the leakage signal.
Example 5
This embodiment is based on any of the above method embodiments, except that the training and testing of the data set is performed in a one-dimensional large kernel residual convolutional neural network and compared with the classification results of other network models.
In this embodiment, the evaluation index is set as:
wherein TP, TN, FP, FN is the abbreviation for true positive, true negative, false positive and false negative, respectively.
And (3) label setting:
different labels are set for different leakage columns to leak. The method is characterized in that the method comprises the steps of defining a C1 when a No. 2 column leaks, setting ten grating area labels to 0, locating the C2 when a No. 3 column leaks, setting the ten grating area labels to 1, defining the C3 when a No. 5 column leaks, setting the ten grating area labels to 2, defining the C4 when a No. 7 column leaks, and setting the ten grating area labels to 2. The label settings are shown in table 1.
Table 1 tag settings
Experimental results: after training the data using the proposed large-kernel residual convolutional neural network, model performance was checked on the test set, and compared using AlexNet and res net networks, and compared using spectrum weighted and large-kernel residual convolutional models without spectrum weighting, model performance versus table 2.
Table 2 model performance comparison
From the data in table 2, it can be observed that the spectrum weighting method and the large-kernel residual convolution neural network model used in the invention are superior to the common neural network model, and have high recognition rate for gas pipeline leakage positioning.
The present application also provides a computer readable storage medium, such as a flash memory, a hard disk, a multimedia card, a card memory (e.g., SD or DX memory, etc.), a Random Access Memory (RAM), a Static Random Access Memory (SRAM), a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), a programmable read-only memory (PROM), a magnetic memory, a magnetic disk, an optical disk, a server, an App application store, etc., having stored thereon a computer program that when executed by a processor performs a corresponding function. The computer readable storage medium of the present embodiment, when executed by a processor, implements the pipeline leakage localization method based on the spectrum weighting and residual convolution neural network of the method embodiment.
It should be noted that each step/component described in the present application may be split into more steps/components, or two or more steps/components or part of the operations of the steps/components may be combined into new steps/components, as needed for implementation, to achieve the object of the present invention.
The sequence numbers of the steps in the above embodiments do not mean the order of execution, and the execution order of the processes should be determined by the functions and the internal logic, and should not be construed as limiting the implementation process of the embodiments of the present application.
It will be understood that modifications and variations will be apparent to those skilled in the art from the foregoing description, and it is intended that all such modifications and variations be included within the scope of the following claims.
Claims (10)
1. The pipeline leakage positioning method based on the spectrum weighting and residual convolution neural network is characterized by comprising the following steps of:
s1, collecting vibration signals in a grating array vibration sensing optical cable, wherein the grating array vibration sensing optical cable is paved above a pipeline in a linear paving mode;
s2, extracting a main frequency in the vibration signal, and carrying out frequency spectrum weighting processing in a certain frequency range by taking the main frequency as a center to obtain a reinforced vibration signal;
s3, inputting the reinforced vibration signals into a trained one-dimensional large-kernel residual convolution neural network model to obtain the pipeline leakage type probability and position the leakage position;
the training process of the one-dimensional large-kernel residual convolution neural network model is as follows: dividing a grating array vibration sensing optical cable paved above a pipeline into a plurality of areas, and arranging leakage holes with a plurality of apertures on the pipeline, wherein leakage in different working conditions is simulated by using different pressures; collecting vibration signals generated by leakage of pipelines of the grating array vibration sensing optical cable under different pressures and different apertures; extracting the main frequency of each sensing area when the pipeline leaks, calibrating the main frequency with the largest occurrence frequency as a leakage frequency, and carrying out frequency spectrum weighting processing in a certain frequency range by taking the leakage frequency as a center to obtain a processed leakage vibration signal as a data sample set; dividing a data sample set into a training set and a testing set, training a pre-constructed one-dimensional large-core residual error convolutional neural network model by adopting the training set, and testing the performance of the one-dimensional large-core residual error convolutional neural network model by adopting the testing set.
2. The method for positioning the leakage of the pipeline based on the spectrum weighted and residual convolution neural network according to claim 1, wherein three different pressures are specifically used, two different leakage holes are used for simulating different working conditions, and a plurality of leakage columns on the pipeline are positioned.
3. The method for locating pipeline leakage based on spectrum weighted and residual convolutional neural network according to claim 1, wherein when the vibration signals in the data sample set are collected, firstly, a high-frequency instrument is used for collecting vibration signals generated by leakage, if the collected signals find that the vibration signals generated by the pipeline leakage are all low-frequency signals, the pipeline leakage signals are subjected to downsampling processing, and the sampling signals are reduced from high frequency to low frequency.
4. The method for locating a pipeline leakage based on a spectrum weighted and residual convolutional neural network according to claim 1, wherein the spectrum weighted processing is performed in a range of 1Hz with the leakage frequency as the center.
5. The method for locating a pipeline leakage based on a spectrum weighted sum residual convolutional neural network according to any one of claims 1-4, wherein the architecture of the one-dimensional large kernel residual convolutional neural network model is:
the first layer is a stage block layer, and the leakage data firstly extracts signal characteristics through a plurality of convolution layers of the stage block layer;
the second layer is a Large Kernel layer and a Small Kernel layer, the output signal features of the first layer respectively enter the Large Kernel layer to extract the global features of the vibration signals, enter the Small Kernel layer to extract the texture features of the vibration signals, and are connected with the signal features of the first layer in a residual way after the global features and the texture features are added;
the third layer, the fourth layer and the second layer have the same structure and are used for extracting deep complex features to form multi-level feature expression;
the fifth layer is a full-connection layer, and the output signal characteristics of the fourth layer are flattened by using a linear layer;
the sixth layer is a softmax layer, and the flattened signal features are classified and output for pipeline leakage using a softmax function.
6. The method for locating pipeline leakage based on spectrum weighted and residual convolutional neural network according to claim 1, wherein the one-dimensional large-kernel residual convolutional neural network model adopts a ReLU activation function, and the loss function adopts a cross entropy function.
7. The method for locating pipeline leakage based on spectrum weighted and residual convolutional neural network of claim 1, wherein the convolution Kernel in the Large Kernel layer is larger than the convolution Kernel in the Small Kernel layer.
8. The method for locating pipeline leakage based on spectrum weighted and residual convolutional neural network according to claim 1, wherein the loss function of the one-dimensional large-kernel residual convolutional neural network model adopts a cross entropy function:
wherein m represents the number of samples participating in network training at a time, y (i) Representing the true probability that the i-th sample is of the leakage class,the i-th sample is represented as the leakage class probability calculated by model training.
9. A pipeline leakage localization system based on a spectrum weighted and residual convolutional neural network, comprising the steps of:
the signal acquisition module is used for acquiring vibration signals in the grating array vibration sensing optical cable, and the grating array vibration sensing optical cable is paved above the pipeline in a linear paving mode;
the signal processing module is used for extracting the main frequency in the vibration signal, and carrying out frequency spectrum weighting processing in a certain frequency range by taking the main frequency as a center to obtain a reinforced vibration signal;
the identification module is used for inputting the reinforced vibration signals into a trained one-dimensional large-kernel residual convolution neural network model to obtain the pipeline leakage type probability and position the leakage position;
the training process of the one-dimensional large-kernel residual convolution neural network model is as follows: dividing a grating array vibration sensing optical cable paved above a pipeline into a plurality of areas, and arranging leakage holes with a plurality of apertures on the pipeline, wherein leakage in different working conditions is simulated by using different pressures; collecting vibration signals generated by leakage of pipelines of the grating array vibration sensing optical cable under different pressures and different apertures; extracting the main frequency of each sensing area when the pipeline leaks, calibrating the main frequency with the largest occurrence frequency as a leakage frequency, and carrying out frequency spectrum weighting processing in a certain frequency range by taking the leakage frequency as a center to obtain a processed leakage vibration signal as a data sample set; dividing a data sample set into a training set and a testing set, training a pre-constructed one-dimensional large-core residual error convolutional neural network model by adopting the training set, and testing the performance of the one-dimensional large-core residual error convolutional neural network model by adopting the testing set.
10. A computer storage medium having stored therein a computer program executable by a processor for performing the method for locating pipeline leakage based on spectral weighting and residual convolution neural network of claim 1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311579219.7A CN117633588A (en) | 2023-11-22 | 2023-11-22 | Pipeline leakage positioning method based on spectrum weighting and residual convolution neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311579219.7A CN117633588A (en) | 2023-11-22 | 2023-11-22 | Pipeline leakage positioning method based on spectrum weighting and residual convolution neural network |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN117633588A true CN117633588A (en) | 2024-03-01 |
Family
ID=90024613
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311579219.7A Pending CN117633588A (en) | 2023-11-22 | 2023-11-22 | Pipeline leakage positioning method based on spectrum weighting and residual convolution neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117633588A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117949687A (en) * | 2024-03-27 | 2024-04-30 | 山东省科学院激光研究所 | Wind speed measurement method and system based on distributed optical fiber sensing and deep learning |
-
2023
- 2023-11-22 CN CN202311579219.7A patent/CN117633588A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117949687A (en) * | 2024-03-27 | 2024-04-30 | 山东省科学院激光研究所 | Wind speed measurement method and system based on distributed optical fiber sensing and deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110059601B (en) | Intelligent fault diagnosis method for multi-feature extraction and fusion | |
| CN111238814B (en) | Rolling bearing fault diagnosis method based on short-time Hilbert transform | |
| CN114429156B (en) | Radar interference multi-domain characteristic countermeasure learning and detection recognition method | |
| CN117633588A (en) | Pipeline leakage positioning method based on spectrum weighting and residual convolution neural network | |
| CN111751133A (en) | Intelligent fault diagnosis method of deep convolutional neural network model based on non-local mean embedding | |
| CN113076920B (en) | Intelligent fault diagnosis method based on asymmetric domain confrontation self-adaptive model | |
| CN111027260A (en) | Reciprocating compressor fault diagnosis method based on neural network | |
| CN114152825A (en) | Fault diagnosis method and device of transformer and fault diagnosis system of transformer | |
| Chen et al. | Automatic modulation classification of radar signals utilizing X-net | |
| CN113537152A (en) | Flow field state fault detection method based on deep neural network | |
| CN115330697A (en) | Tire flaw detection domain self-adaption method based on migratable Swin transducer | |
| CN109782158B (en) | Analog circuit diagnosis method based on multi-stage classification | |
| CN111983569B (en) | Radar interference suppression method based on neural network | |
| CN117473414A (en) | Bearing fault position identification method based on low-noise time-frequency image | |
| CN115356599B (en) | Multi-mode urban power grid fault diagnosis method and system | |
| Ou et al. | Research on Rolling Bearing Fault Diagnosis Using Improved Majorization‐Minimization‐Based Total Variation and Empirical Wavelet Transform | |
| CN108664936B (en) | Diagnosis method and system based on machine fault | |
| CN116910648A (en) | GIS equipment partial discharge spectrum analysis method, system and medium | |
| CN116541771A (en) | Unbalanced sample bearing fault diagnosis method based on multi-scale feature fusion | |
| CN116399592A (en) | Bearing fault diagnosis method based on channel attention dual-path feature extraction | |
| CN116610990A (en) | Method and device for identifying hidden danger of breaker based on characteristic space differentiation | |
| CN115982566A (en) | Multi-channel fault diagnosis method for hydroelectric generating set | |
| CN112529080B (en) | Image generation method based on spectrum feature discrimination | |
| CN115326396A (en) | Bearing fault diagnosis method and device | |
| CN113221736A (en) | Electrocardio identity recognition method and system based on multi-feature collaborative non-negative matrix factorization |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication |